Farming systems are complex and have several dimensions that interact in a dynamic and continuous manner depending on farmers' management strategies. This complexity peaks in Indian semi-arid regions, where small farms encounter a highly competitive environment for markets and resources, especially unreliable access to water from rainfall and irrigation. NAMASTE, a dynamic computer model for water management at the farm level, was developed to reproduce interactions between decisions (investment and technical) and processes (resource management and biophysical) under scenarios of climate-change, socio-economic and water-management policies. The most relevant and novel aspects are i) system-based representation of farming systems, ii) description of dynamic processes via management flexibility and adaptation, iii) representation of farmers' decision-making processes at multiple temporal and spatial scales, iv) management of shared resources. NAMASTE's ability to simulate farmers' adaptive decision-making processes is illustrated by simulating a virtual Indian village composed of two virtual farms with access to groundwater.

Developing sustainable crop systems is a major challenge. Presently, management practices are simulated using either biophysical models or simple farmer decision models. As a result, there is a lack of generic models integrating both biophysical parameters and farmer decision parameters. Here, we developed an original graphical plug-in to sketch and implement decision-making models and to link them with biophysical models. For that, we used the RECORD platform, standing for REnovation and COORDination of agro-ecosystem modeling. Different pop-up windows allow to create the model using a decision formalism then to implement the model under the RECORD platform. The sequence of technical operations is formally modeled as a direct multi-graph without retroaction. The plug-in allows defining activities, relation between activities, and decision rules to trigger the different activities. The resulting model is independent of any biophysical model and can then be linked with different crop models. An example is given on an innovative cropping systems part of the MicMac-Design project. The decision-making model is then linked with the STICS crop model.

RECORD is a modeling and simulation platform with VLE as the main software that combines different forms of equations to represent agrosystem functioning. Although RECORD has a graphic user interface based on box-and-arrow diagrams to develop models, members of modeling communities who prefer Forrester diagrams may become disoriented when using RECORD. To help them, we developed a plugin that displays models as Forrester diagrams and translates them into DEVS formalism.

When dealing with high dimensional and low sample size data, feature selection is often needed to help reduce the dimension of the variable space while optimizing the classification task. Few tools exist for selecting variables in such data sets, especially when classes are numerous ( > 2). We have developed ofw, an R package that implements, in the context of classification, the meta algorithm "optimal feature weighting". We focus on microarray data, although the method can be applied to any p >> n problems with continuous variables. The aim is to select relevant variables and to numerically evaluate the resulting variable selection. Two versions are proposed with the application of supervised multiclass classifiers such as classification and regression trees and support vector machines. Furthermore, a weighted approach can be chosen to deal with unbalanced multiclasses, a common characteristic in microarray data sets.

Radiation hybrid (RH) mapping is a somatic cell technique that is used for ordering markers along a chromosome and estimating physical distances between them. It nicely complements the genetic mapping technique, allowing for finer resolution. Like genetic mapping, RH mapping consists in finding a marker ordering that maximizes a given criteria. Several software packages have been recently proposed to solve RH mapping problems. Each package offers specific criteria and specific ordering techniques. The most general packages look for maximum likelihood maps and may cope with errors, unknowns and polyploid hybrids at the cost of limited computational efficiency. More efficient packages look for minimum breaks or two-points approximated maximum likelihood maps but ignore errors, unknowns and polyploid hybrids. In this paper, we present a simple improvement of the EM algorithm [5] that makes maximum likelihood estimation much more efficient (in practice and to some extent in theory too). The boosted EM algorithm can deal with unknowns in both error-free haploid data and error-free backcross data. Unknowns are usually quite limited in RH mapping but cannot be ignored when one deals with genetic data or multiple populations/panels consensus mapping (markers being not necessarily typed in all panels/populations). These improved EM algorithms have been implemented in the CARTHAGÈNE software. We conclude with a comparison with similar packages (RHMAP and MapMaker) using simulated data sets and present preliminary results on mixed simultaneous RH/genetic mapping on pig data.

Nearly 7000 Arabidopsis thaliana-expressed sequence tags (ESTs) from 10 cDNA libraries have been sequenced, of which almost 5000 non-redundant tags have been submitted to the EMBL data bank. The quality of the cDNA libraries used is analysed. Similarity searches in international protein data banks have allowed the detection of significant similarities to a wide range of proteins from many organisms. Alignment with ESTs from the rice systematic sequencing project has allowed the detection of amino acid motifs which are conserved between the two organisms, thus identifying tags to genes encoding highly conserved proteins. These genes are candidates for a common framework in genome mapping projects in different plants.